Skip to contents

Non-Negative PCA with Deflation (nPCA)

npca.Rd

Computes a non-negative PCA (nPCA) basis via iterative regression and deflation on an input matrix x.

Usage

npca(x, k, tol = 1e-10, max.iter = 1000, return.score = TRUE)

Arguments

x

A numeric matrix of size (p x n). The row dimension will be embedded

k

Integer; number of nPCA components to extract.

tol

Numeric; convergence tolerance for nPCA iterations. Default: 1e-10.

max.iter

Integer; maximum iterations for calculating each nPCA component. Default: 1000.

return.score

Logical; If TRUE return the nPCA projection of x.

Value

A list with:

  • scores: Matrix (n x k) of non-negative nPCA scores (columns are components).

  • loadings: Matrix (p x k) of non-negative nPCA loadings

  • sd: Numeric vector (k) of nPCA “standard deviations”.

Details

This function implements non-negative PCA using restricted single-value decomposition, which is solved using the iterative regression approach introduced by Sigg and Buhmann (2008) and deflation approach introduced by Mackey (2008).

References

Sigg, C. D., & Buhmann, J. M. (2008, July). Expectation-maximization for sparse and non-negative PCA. In Proceedings of the 25th international conference on Machine learning (pp. 960-967).

Mackey, L. (2008). Deflation methods for sparse PCA. Advances in neural information processing systems, 21.

Examples

x <- matrix(rnorm(1000), nrow - 100)
#> Error in nrow - 100: non-numeric argument to binary operator
res <- nn.pca(x, k = 10, tol = 1e-10, n.iter = 1000)
#> Error in nn.pca(x, k = 10, tol = 1e-10, n.iter = 1000): could not find function "nn.pca"
str(res$loadings); str(res$sd)
#> Error: object 'res' not found